Enhance the Concrete Crack Classification Based on a Novel Multi-Stage YOLOV10-ViT Framework DOI Creative Commons
Ali Mayya, Nizar Faisal Alkayem

Sensors, Год журнала: 2024, Номер 24(24), С. 8095 - 8095

Опубликована: Дек. 19, 2024

Early identification of concrete cracks and multi-class detection can help to avoid future deformation or collapse in structures. Available traditional methodologies require enormous effort time. To overcome such difficulties, current vision-based deep learning models effectively detect classify various cracks. This study introduces a novel multi-stage framework for crack type classification. First, the recently developed YOLOV10 model is trained possible defective regions images. After that, modified vision transformer (ViT) images into three main types: normal, simple cracks, multi-branched The evaluation process includes feeding test model, identifying defect regions, finally delivering detected ViT which decides appropriate those regions. Experiments are conducted using individual proposed framework. improve generation ability, multi-source datasets structures used. For classification part, dataset consisting 12,000 classes utilized, while composed materials from historical buildings containing 1116 with their corresponding bounding boxes, utilized. Results prove that accurately classifies types 90.67% precision, 90.03% recall, 90.34% F1-score. results also show outperforms by 10.9%, 19.99%, 19.2% F1-score, respectively. YOLOV10-ViT be integrated construction systems based on obtain early warning

Язык: Английский

Triple-stage crack detection in stone masonry using YOLO-ensemble, MobileNetV2U-net, and spectral clustering DOI
Ali Mayya, Nizar Faisal Alkayem

Automation in Construction, Год журнала: 2025, Номер 172, С. 106045 - 106045

Опубликована: Фев. 7, 2025

Язык: Английский

Процитировано

2
Bilingual Sign Language Recognition: A YOLOv11-Based Model for Bangla and English Alphabets DOI Creative Commons

Nawshin Navin,

Fahmid Al Farid,

Raiyen Z. Rakin

и другие.

Journal of Imaging, Год журнала: 2025, Номер 11(5), С. 134 - 134

Опубликована: Апрель 27, 2025

Communication through sign language effectively helps both hearing- and speaking-impaired individuals connect. However, there are problems with the interlingual communication between Bangla Sign Language (BdSL) English (ASL) due to absence of a unified system. This study aims introduce detection system that incorporates these two languages enhance flow for those who use forms language. developed tested deep learning-based sign-language can recognize BdSL ASL alphabets concurrently in real time. The approach uses YOLOv11 object architecture has been trained an open-source dataset on set 9556 images containing 64 different letter signs from languages. Data preprocessing was applied performance model. Evaluation criteria, including precision, recall, mAP, other parameter values were also computed evaluate analysis proposed method shows precision 99.12% average recall rates 99.63% 30 epochs. studies show model outperforms current techniques recognition (SLR) be used communicating assistive technologies human-computer interaction systems.

Язык: Английский

Процитировано

1
YOLOv9s-Pear: A Lightweight YOLOv9s-Based Improved Model for Young Red Pear Small-Target Recognition DOI Creative Commons
Yi Shi,

Zhen Duan,

Shunhao Qing

и другие.

Agronomy, Год журнала: 2024, Номер 14(9), С. 2086 - 2086

Опубликована: Сен. 12, 2024

With the advancement of computer vision technology, demand for fruit recognition in agricultural automation is increasing. To improve accuracy and efficiency recognizing young red pears, this study proposes an improved model based on lightweight YOLOv9s, termed YOLOv9s-Pear. By constructing a feature-rich diverse image dataset pears introducing spatial-channel decoupled downsampling (SCDown), C2FUIBELAN, YOLOv10 detection head (v10detect) modules, YOLOv9s was enhanced to achieve efficient small targets resource-constrained environments. Images were captured at different times locations underwent preprocessing establish high-quality dataset. For improvements, integrated general inverted bottleneck blocks from C2f MobileNetV4 with RepNCSPELAN4 module form new C2FUIBELAN module, enhancing model’s training speed small-scale object detection. Additionally, SCDown v10detect modules replaced original AConv structures model, further improving performance. The experimental results demonstrated that YOLOv9s-Pear achieved high while reducing computational costs parameters. accuracy, recall, mean precision, extended precision 0.971, 0.970, 0.991, 0.848, respectively. These confirm SCDown, pear tasks. findings not only provide fast accurate technique but also offer reference detecting fruits other trees, significantly contributing technology.

Язык: Английский

Процитировано

9
Performance Evaluation of Deep Learning Image Classification Modules in the MUN-ABSAI Ice Risk Management Architecture DOI Creative Commons
Ravindu G. Thalagala, Oscar De Silva,

Dan Oldford

и другие.

Sensors, Год журнала: 2025, Номер 25(2), С. 326 - 326

Опубликована: Янв. 8, 2025

The retreat of Arctic sea ice has opened new maritime routes, offering faster shipping opportunities; however, these routes present significant navigational challenges due to the harsh conditions. To address challenges, this paper proposes a deep learning-based risk management architecture with multiple modules, including classification, assessment, floe tracking, and load calculations. A comprehensive dataset 15,000 images was created using public sources contributions from Canadian Coast Guard, it used support development evaluation system. performance YOLOv8n-cls model assessed for classification modules its fast inference speed, making suitable resource-constrained onboard systems. training were conducted across platforms, Roboflow, Google Colab, Compute Canada, allowing detailed comparison their capabilities in image preprocessing, training, real-time generation. results demonstrate that Image Classification Module I achieved validation accuracy 99.4%, while II attained 98.6%. Inference times found be less than 1 s Colab under 3 on stand-alone system, confirming architecture's efficiency condition monitoring.

Язык: Английский

Процитировано

0
Real-Time American Sign Language Interpretation Using Deep Learning and Keypoint Tracking DOI Creative Commons
Bader Alsharif, Easa Alalwany, Ali Ibrahim

и другие.

Sensors, Год журнала: 2025, Номер 25(7), С. 2138 - 2138

Опубликована: Март 28, 2025

Communication barriers pose significant challenges for the Deaf and Hard-of-Hearing (DHH) community, limiting their access to essential services, social interactions, professional opportunities. To bridge this gap, assistive technologies leveraging artificial intelligence (AI) deep learning have gained prominence. This study presents a real-time American Sign Language (ASL) interpretation system that integrates with keypoint tracking enhance accessibility foster inclusivity. By combining YOLOv11 model gesture recognition MediaPipe precise hand tracking, achieves high accuracy in identifying ASL alphabet letters real time. The proposed approach addresses such as ambiguity, environmental variations, computational efficiency. Additionally, enables users spell out names locations, further improving its practical applications. Experimental results demonstrate attains mean Average Precision ([email protected]) of 98.2%, an inference speed optimized real-world deployment. research underscores critical role AI-driven empowering DHH community by enabling seamless communication interaction.

Язык: Английский

Процитировано

0
Enhanced Vehicle Identification Using YOLOv8 with Counter-Based Grouping for Improved Real-Time Performance DOI
Ankit Agrawal,

C. K. Shukla

Algorithms for intelligent systems, Год журнала: 2025, Номер unknown, С. 1 - 13

Опубликована: Янв. 1, 2025

Язык: Английский

Процитировано

0
Small Object Detection in UAV Remote Sensing Images Based on Intra-Group Multi-Scale Fusion Attention and Adaptive Weighted Feature Fusion Mechanism DOI Creative Commons
Zhe Yuan,

Jianglei Gong,

Baolong Guo

и другие.

Remote Sensing, Год журнала: 2024, Номер 16(22), С. 4265 - 4265

Опубликована: Ноя. 15, 2024

In view of the issues missed and false detections encountered in small object detection for UAV remote sensing images, inadequacy existing algorithms terms complexity generalization ability, we propose a model named IA-YOLOv8 this paper. This integrates intra-group multi-scale fusion attention mechanism adaptive weighted feature approach. extraction phase, employs hybrid pooling strategy that combines Avg Max to replace single operation used original SPPF framework. Such modifications enhance model’s ability capture minute features objects. addition, an module is introduced, which capable automatically adjusting weights based on significance contribution at different scales improve sensitivity Simultaneously, lightweight implemented, aims effectively mitigate background interference saliency Experimental results indicate proposed has parameter quantity 10.9 MB, attaining average precision (mAP) value 42.1% Visdrone2019 test set, mAP 82.3% DIOR 39.8% AI-TOD set. All these outperform algorithms, demonstrating superior performance task sensing.

Язык: Английский

Процитировано

1
From Stationary to Nonstationary UAVs: Deep-Learning-Based Method for Vehicle Speed Estimation DOI Creative Commons
Muhammad Waqas Ahmed, Muhammad Adnan, Muhammad Ahmed

и другие.

Algorithms, Год журнала: 2024, Номер 17(12), С. 558 - 558

Опубликована: Дек. 6, 2024

The development of smart cities relies on the implementation cutting-edge technologies. Unmanned aerial vehicles (UAVs) and deep learning (DL) models are examples such disruptive technologies with diverse industrial applications that gaining traction. When it comes to road traffic monitoring systems (RTMs), combination UAVs vision-based methods has shown great potential. Currently, most solutions focus analyzing footage captured by hovering due inherent georeferencing challenges in video from nonstationary drones. We propose an innovative method capable estimating speed using both stationary UAVs. process involves matching each pixel input frame a georeferenced orthomosaic feature-matching algorithm. Subsequently, tracking-enabled YOLOv8 object detection model is applied detect their trajectories. geographic positions these moving over time logged JSON format. accuracy this was validated reference measurements recorded laser gun. results indicate proposed can estimate vehicle speeds absolute error as low 0.53 km/h. study also discusses associated problems constraints drone proposes strategies for minimizing noise inaccuracies. Despite challenges, framework demonstrates considerable potential signifies another step towards automated systems. This system enables transportation modelers realistically capture behavior wider area, unlike existing roadside camera prone blind spots limited spatial coverage.

Язык: Английский

Процитировано

1
Enhance the Concrete Crack Classification Based on a Novel Multi-Stage YOLOV10-ViT Framework DOI Creative Commons
Ali Mayya, Nizar Faisal Alkayem

Sensors, Год журнала: 2024, Номер 24(24), С. 8095 - 8095

Опубликована: Дек. 19, 2024

Early identification of concrete cracks and multi-class detection can help to avoid future deformation or collapse in structures. Available traditional methodologies require enormous effort time. To overcome such difficulties, current vision-based deep learning models effectively detect classify various cracks. This study introduces a novel multi-stage framework for crack type classification. First, the recently developed YOLOV10 model is trained possible defective regions images. After that, modified vision transformer (ViT) images into three main types: normal, simple cracks, multi-branched The evaluation process includes feeding test model, identifying defect regions, finally delivering detected ViT which decides appropriate those regions. Experiments are conducted using individual proposed framework. improve generation ability, multi-source datasets structures used. For classification part, dataset consisting 12,000 classes utilized, while composed materials from historical buildings containing 1116 with their corresponding bounding boxes, utilized. Results prove that accurately classifies types 90.67% precision, 90.03% recall, 90.34% F1-score. results also show outperforms by 10.9%, 19.99%, 19.2% F1-score, respectively. YOLOV10-ViT be integrated construction systems based on obtain early warning

Язык: Английский

Процитировано

0